摘要
目的:探究甲状腺结节周围区域超声影像组学特征预测甲状腺乳头状癌(PTC)颈部淋巴结转移(LNM)的效能,并基于影像组学特征构建预测模型验证其性能。方法:回顾性收集2021年3月至2024年6月于陕西省中医医院收治并经甲状腺手术后病理证实的237例PTC患者,将其按7∶3比例分为训练集(166例)和验证集(71例)。根据术后病理将237例患者分为转移组(108例)和非转移组(129例)。收集所有患者临床病历资料及常规超声特征信息;通过定量提取、筛选结节内和结节周围区域的超声影像组学特征,利用机器学习分类器构建影像组学特征模型,得到影像组学特征评分(Rad-Score),对比结节内、结节周围Rad-Score及转移组和非转移组结节周围的Rad-Score。于训练集分析影响颈部LNM的独立危险因素,构建临床-超声模型,并结合Rad-Score构建基于结节周围影像组学特征的联合模型,通过受试者工作特征(ROC)曲线分析对比其预测效能;构建联合模型列线图,经校准曲线和Hosmer-Lemeshow检验评估列线图的校准度和拟合度。结果:在训练集中,通过最小绝对收缩选择算子(LASSO)算法,分别提取并筛选出6个结节内影像组学特征和11个结节周围影像组学特征。转移组结节周围Rad-Score分别为(7.43±0.45)和(7.19±0.51)分,明显高于非转移组结节周围的(3.25±0.28)和(3.51±0.32)分,差异均有统计学意义(t=72.708,61.222,P<0.05)。年龄、包膜侵犯、微钙化、超声提示淋巴结阳性及结节周围Rad-Score是影响PTC患者颈部LNM的独立危险因素(OR=0.592、2.983、3.593、4.424、2.575,P<0.05)。ROC曲线显示,联合模型在训练集和验证集中预测LNM的ROC曲线下面积(AUC)值分别为0.861、0.872,优于临床-超声模型的0.759、0.783。结论:在训练集和验证集中,联合模型列线图预测PTC患者颈部LNM均具有良好的校准度和拟合度。基于结节周围超声影像组学特征构建的临床模型对术前预测PTC颈部淋巴结状态具有良好的效能,有望成为个体化预测LNM的有效工具。
Objective:To explore the efficacy of the features of ultrasound imaging omics at the area peripheral thyroid nodule in predicting cervical lymph node metastasis(LNM)of papillary thyroid carcinoma(PTC),and construct a prediction model based on the features of imaging omics and to verify its performance.Methods:A total of 237 PTC patients who admitted to Shaanxi Provincial Hospital of Chinese Medicine from March 2021 to June 2024 and were confirmed by pathology were retrospectively collected.They were divided into the training set(166 cases)and the validation set(71 cases)as a ratio of 7 to 3.According to the postoperatively pathological results,237 patients were divided into the metastasis group(108 cases)and the non-metastasis group(129 cases).The clinical data and conventional ultrasound characteristic information of all patients were collected,and a feature model of imaging omics was constructed through quantitative extracting and screening the features of ultrasound imaging omics within nodules and peripheral nodules,and utilizing machine learning classifier.Then,the feature score(Rad-Score)of image omics was obtained.The Rad-Score values within and peripheral nodules,and the Rad-Score values peripheral nodules of metastasis group and non-metastasis group were compared.In training set,the independent risk factors of affecting neck LNM were analyzed,and a clinical-ultrasonic model was constructed,which was combined with Rad-Score to construct a joint model based on the features of imaging omics peripheral nodules.The receiver operating characteristic(ROC)curve was used to analyze and compare its predictive efficacy.The nomogram of the joint model was constructed,and then,the calibration and fitting degrees of the nomogram were assessed by calibration curve and Hosmer-Lemeshow test.Result:In training set,6 features of imaging omics within nodules and 11 features of imaging omics peripheral nodules were respectively extracted and screened out through the least absolute shrinkage and selection operator(LASSO)algorithm.In the training set and validation set,the Rad-Scores peripheral the nodules in the metastasis group were respectively(7.43±0.45)points and(7.19±0.51)points,which were significantly higher than(3.25±0.28)points and(3.51±0.32)points peripheral the nodules in the non-metastasis group,and the differences were statistically significant(t=72.708,61.222,P<0.05).The results of factor analysis showed that age,capsule invasion,microcalcification,ultrasound-indicated lymph node positivity and Rad-Score around nodules were independent risk factors of affecting cervical LNM of PTC patients(OR=0.592,2.983,3.593,4.424,2.575,P<0.05).The ROC curve showed that the area under curve(AUC)values of the ROC curve of joint model in training set and validation set were respectively 0.861 and 0.872 in predicting LNM,respectively,which were superior to 0.759 and 0.783 of the clinical-ultrasound model.Conclusion:In both the training set and the validation set,the nomogram of joint model has favorable calibration and fitting in predicting cervical LNM of PTC patients.The construction of clinical model based on the features of ultrasound imaging omics peripheral nodules has a favorable efficacy in predicting the status of cervical lymph node of PTC patients before surgery,which is expected to be an effective tool of individual prediction for LNM.
作者
杨金艳
方媛
张开元
强文思
张新妹
Yang Jinyan;Fang Yuan;Zhang Kaiyuan;Qiang Wensi;Zhang Xinmei(Department of Ultrasound,Shaanxi Provincial Hospital of Chinese Medicine,Xi'an 710000,China;Department of Ultrasound Medicine,Xi'an Honghui Hospital affiliated to Xi'an Jiaotong University,Xi'an 730000,China;Department of Imaging,The First Affiliated Hospital of Xi'an Medical University,Xi'an 710077,China)
出处
《中国医学装备》
2025年第11期74-80,共7页
China Medical Equipment
基金
国家自然科学基金青年科学基金(81903072)
陕西省重点研发计划(2023SF-107)
陕西省自然科学基础研究计划杰出青年科学基金(2023-JC-JQ-170)。
